Dear FSLers
Need some advice about Feat statistics set up for a within subject Feat analysis examining effect of an intervention causing weight loss.
Subjects scanned at baseline and after an intervention (pre and post visits).
Have done standard paired t-test to look at effects of intervention (pre-post and post-pre contrasts), including EVs to identify the paired subjects with all pre scans and all post scans entered as data.
Now want to look at how changes in body weight, or a behavioural variables e.g. appetite rating, between the 2 time points relates to changes in BOLD signal.
Note that everybody lost weight and for most other variables the majority show a decrease between the 2 time points.
Questions:
1. If I include an additional co-variate (whether body weight or appetite) as an extra EV using the demeaned value at each time point over the whole group (pre or post), and then have 1 or -1 as the contrast for positive or negative correlation with body weight/appetite, do I use this instead of (or as well as) the first EV that is generated as a visit variable (-1 or 1).
If I do not replace this visit EV with the demeaned co-variate EV, then the co-variate EV will likely not be orthogonal to the visit EV, especially as everyone loses weight and the majority reduce appetite.
Would this be different if on average the co-variate did not change significantly between the 2 visits?
2. If I wanted to look at correlation with % weight loss, do I complete the EV as 0 for all the baseline pre visits, and demeaned % weight loss for the post visits?
3. If the co-variate is only measured at baseline or does not change (e.g. gender) can I simply re-enter the variable for both pre and post-visits, but this time including the visit EV?
4. Or should I instead for any (or some of the) analyses above, subtract the post scan cope.nii.gz from the pre scan cope.nii.gz using fslmaths and then just have a single co-variate (with no need for any pairing) to correlate BOLD with the delta in the co-variate, or baseline co-variate value (in case of 3).
Many thanks
Tony Goldstone
Division of Brain Sciences, Imperial College London, UK
|